Building a machine to make the real work faster

Choosing to stop the project and build an instrument instead.

Building a machine to make the real work faster is worth doing whenever the time a repeated task will cost, multiplied across every time it needs doing, comfortably exceeds the time it takes to build something that does the task properly. Stated plainly that sounds obvious, and it is nonetheless routinely skipped by anyone under enough pressure to feel that stopping to build a tool is itself a delay the schedule cannot afford.

Cutting foam shapes by hand took three days each. Building a machine to do it took two, and the only reason that machine got built at all was sitting down and doing that arithmetic instead of trusting the feeling that another three-day cut was the faster, safer choice simply because it was the familiar one.

Why the familiar task always feels cheaper

A repeated manual task feels cheap because each instance is a known quantity, a predictable few days of familiar work with no risk of the setbacks a new build might run into. Building a tool is unfamiliar, can overrun, and delivers nothing useful until it is finished, so it feels expensive by comparison even when the arithmetic says otherwise. That asymmetry keeps teams grinding through a repeated task by hand long after a proper tool would have paid for itself.

The way the two costs are paid makes it worse. The manual cost arrives in small, tolerable instalments, a few days here and a few days there, never large enough at any single moment to force a stop and a proper accounting. The tool-building cost is one visible lump paid before anything comes back, and the two feel wildly different in size even when the running total favours the tool.

A blunt axe and an afternoon of firewood

Spending twenty minutes sharpening a blunt axe before an afternoon of chopping firewood feels like twenty minutes not spent chopping, and it is tempting to skip it and swing the blunt axe harder. The sharpened axe pays that twenty minutes back within the first few logs and keeps paying with every log afterward, while the blunt one costs a little extra effort on every swing for as long as the chopping lasts. Building a machine before the real work is the same trade at a larger scale.

The foam machine was an unusually lopsided case of it. At two days to build against three days per manual cut, it had paid for itself before a single further cut by hand would have been finished, and every cut after that arrived at the machine's running cost. Most build-versus-repeat decisions are closer, weighing an uncertain number of future repetitions against an uncertain build time, but a lopsided one is worth recognising, because hesitating there costs real time and buys no caution in return.

Counting the repetitions still to come

The decision rests almost entirely on how many more times the task will repeat. A tool built to replace a task that will happen once or twice more rarely pays for itself, and the same tool facing a task expected dozens of times almost always does, so the honest first question is a concrete estimate of how many repetitions the manual method still faces. Getting that estimate roughly right, and early, matters more than getting the machine perfect. A first version that does eighty percent of the job well is worth building the moment the arithmetic favours it, since a mediocre tool started today beats a perfect one started after several more weeks of unrecovered manual work.

The case for building weakens when the remaining repetitions are few, or when the task varies enough from one instance to the next that no single machine could replace all of it. A tool built for a narrow job is often useless once the job's shape changes slightly, and forcing it onto a task that keeps shifting can cost more in rework than the manual approach ever did. Telling a truly repetitive task from a superficially similar but varied one is a skill built mostly by guessing wrong about it at least once. The next two articles look at this machine from the inside, first at a quirk of how it cuts, then at when this kind of detour pays for itself and when it quietly does not.

My key error with this

We were cutting a run of acrylic mounts, and I did the first few the way anyone does the first few, marking each blank out individually and cutting to the line, which works and produces parts that are each very slightly different from one another. The differences did not matter while I was the only person cutting and assembling them, because I knew where each one was meant to sit. They mattered a great deal as soon as the parts were assembled by somebody else, since nothing was interchangeable and every mount had effectively become a bespoke component with no marking to say so. Building a simple fixture that held the blank in one repeatable position fixed the cutting, and the more useful accident was that the same fixture then worked as a template during assembly, showing where the mounts belonged rather than requiring the information to be remembered. What replaced the belief is that the value of a fixture is repeatability rather than speed, and that a jig which also communicates the correct arrangement is doing two jobs for the price of one afternoon.